Case stage 02
Documents
Records arrive from EHRs, laboratories, imaging centers, pharmacies, HIEs, payers, and partners in many formats. Each source is preserved, linked to the workflow and patient or member identity, and disclosed only for an authorized purpose.
Partly builtReference parts exist; the solution layer is designedExtraction, imaging attachments, and role-trimmed retrieval have reference implementations. The Epic connection is designed and is not yet connected to a live Epic environment.
What happens
- Records arrive through FHIR from Epic and other EHR systems, as HL7 v2 messages through the integration engine, as CCD or C-CDA documents, as PDF or fax, and as DICOM imaging exams.
- Each record is validated, identity-matched, and linked to the correct workflow without losing its original source.
- Every read and every share is written to the audit trail.
How the AI platform addresses it
- Scanned and faxed pages are read with Azure Document Intelligence. Each extracted field keeps the page and region it came from.
- Imaging studies are linked through DICOMweb so a study can be found from the encounter, order, claim, or workflow record.
- Access is checked on every read. A reader sees only the records their role allows, and answers are trimmed to those records.
AI role
Field extraction from scans and faxes. Retrieval over records returns only the records the reader may see, with citations.
Approval
The coordinator checks each field that the extraction flags as uncertain. Records are never edited by the model.
Systems involved
- Epic and other EHRs
- Integration engines
- Document intelligence
- Azure Health Data Services
- Document management
Reference components
- Claims document intake
- Medical records extraction flow
- Imaging claims attachments
- Claims document memory with role trimming